This project will investigate neuropsychological and computational models of semantic memory in healthy people and in patients with Alzheimer's disease (AD) or category-specific knowledge deficits. Specifically, this study will focus on two types of semantic knowledge: perceptual and functional. The semantic priming paradigm will be used to study the effects of perceptual and functional similarity on reading time. Experiments 1 and 2 test the hypothesis that differences in the speed of accessibility of perceptual and functional features reflect the degree of intercorrelations between the features; it is predicted that perceptual features will become accessible more quickly than will functional features, both in normal subjects and in simulations of a connectionist model. Experiments 3 and 4 test the hypothesis that patients with AD or focal brain damage who show specific impairments in the knowledge of living things will have impaired access to perceptual features; it is predicted that perceptual but not functional priming will be reduced in these patients, and that this impairment can be simulated by random damage to the connections in the model as a result of the high degree of intercorrelations between perceptual features. Experiment 5 will test the hypothesis that the effects of word frequency will be greater for living things than for non-living things as a result of differences in the informativeness of perceptual and functional features. These differential effects of word frequency should be observed in normal subjects, in AD patients with category-specific knowledge deficits, and in simulations of a connectionist model. Experiments 6, 7, and 8 will study neuroanatomical dissociations of semantic features using functional magnetic resonance imaging (fMRI) to localize regions of increased activity during perceptual and functional processing of words and pictures. It is predicted that regions of increased activation during processing of perceptual attributes will be more posterior in cortex than during processing of functional attributes.